> ## Documentation Index
> Fetch the complete documentation index at: https://docs.sigmora.org/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent skill

> Drive Sigma from Claude Code, Codex or Cursor with the open Sigmora skill and your API key.

The Sigmora skill teaches a coding agent to work with **Sigma**, the co-editor
inside Sigmora: edit footage, cut a talk to a length, add captions or music,
make carousels, generate clips and UGC ads, recreate a reference post, research
ads and trends, and schedule idea runs. It is the same Sigma as the web app and
the chat apps, so every conversation also shows in the project's Sigma panel.

The skill is a `SKILL.md` plus a small Node script (no dependencies) that calls
`/v1/sigma` with your key. If your agent speaks MCP, the same capabilities are
MCP tools; see [MCP](/mcp-server).

## Install

```bash theme={"theme":{"light":"github-light-default","dark":"github-dark-default"}}
npx skills add https://sigmora.org/kit/sigmora-skill.zip
```

Or install it by hand: download
[`sigmora-skill.zip`](https://sigmora.org/kit/sigmora-skill.zip) and unzip it
into a `sigmora` folder in your agent's skills folder, for example:

```bash theme={"theme":{"light":"github-light-default","dark":"github-dark-default"}}
curl -fsSLo sigmora-skill.zip https://sigmora.org/kit/sigmora-skill.zip
unzip sigmora-skill.zip -d ~/.claude/skills/sigmora
```

The skill is MIT licensed.

Then set your key. Mint it in **Settings → API keys**; it acts and bills as
you, in the workspace it belongs to.

```bash theme={"theme":{"light":"github-light-default","dark":"github-dark-default"}}
export SIGMORA_API_KEY=sk_live_...
```

## What the agent does

| Step | Command | Endpoint |
| - | - | - |
| Talk to Sigma; start or continue a thread, in a project or not | `sigmora turn "<message>" [--thread] [--project] [--mode]` | `POST /v1/sigma/turns` |
| Point at one frame (Mark-to-Fix) | `--at <sec> --region x,y,w,h [--still <asset>]` | `mark` on the turn |
| Apply an edit proposed in review mode (free) | `sigmora apply '<proposal>'` | `POST /v1/sigma/proposals/apply` |
| Run paid work, after the user agrees | `sigmora approve '<approval>' --thread <id>` | `POST /v1/sigma/approvals` |
| Follow a render | `sigmora job <jobId>` | `GET /v1/jobs/{jobId}` |
| Read what a project holds | `sigmora project <projectId>` | `GET /v1/projects/{projectId}` |
| Take back agentic edits (free) | `sigmora revert <turnId> ...` | `POST /v1/sigma/revert` |

## Approval is the user's, not the agent's

A turn never spends. Paid work comes back under `approvals[]` with its quote:
the credits, the model that will run, and, when it cannot run yet, the missing
setting by name. The skill tells the agent to show that quote and approve only
after you say yes. Approving is charged only after the work succeeds, and
retrying with the same idempotency key replays the result instead of charging
twice. If your workspace auto-approves small amounts, those approvals come back
already run and marked `autoApproved`.

## Research and scheduled ideas

Ask in plain words: "what ads are competitors running for vitamin C serums?"
Sigma searches the Meta Ad Library, your watched trends and the web, and cites
each source. "Send me 5 ideas every Monday for this project" becomes a schedule
approval that is free to set up; each run is one Sigma reply, billed like one,
and posts into the project's thread (and a linked Slack, Discord, Telegram,
Teams or WhatsApp conversation). Pause or delete schedules on the project's Plan
board or in Settings.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.